Attention U-Net: Learning Where to Look for the Pancreas

Attention U-Net: Learning Where to Look for the Pancreas

20 May 2018 | Ozan Oktay1,5, Jo Schlemper1, Loic Le Folgoc1, Matthew Lee4, Mattias Heinrich3, Kazumari Misawa2, Kensaku Mori2, Steven McDonagh1, Nils Y Hammerla5, Bernhard Kainz1, Ben Glocker1, and Daniel Rueckert1
The paper introduces a novel attention gate (AG) model for medical image segmentation, specifically focusing on the pancreas in abdominal CT scans. The AG model is designed to automatically focus on target structures of varying shapes and sizes, eliminating the need for explicit external tissue/organ localization modules. The proposed Attention U-Net architecture integrates AGs into standard U-Net models, enhancing sensitivity and prediction accuracy while maintaining computational efficiency. Experimental results on two large CT abdominal datasets (TCIA Pancreas CT-S2 and CT-150) demonstrate that AGs consistently improve prediction performance across different datasets and training sizes. The source code for the Attention U-Net is publicly available. The paper also discusses related work in CT pancreas segmentation and attention mechanisms, and provides a detailed methodology for the proposed model, including its integration into the U-Net architecture and evaluation on challenging datasets.The paper introduces a novel attention gate (AG) model for medical image segmentation, specifically focusing on the pancreas in abdominal CT scans. The AG model is designed to automatically focus on target structures of varying shapes and sizes, eliminating the need for explicit external tissue/organ localization modules. The proposed Attention U-Net architecture integrates AGs into standard U-Net models, enhancing sensitivity and prediction accuracy while maintaining computational efficiency. Experimental results on two large CT abdominal datasets (TCIA Pancreas CT-S2 and CT-150) demonstrate that AGs consistently improve prediction performance across different datasets and training sizes. The source code for the Attention U-Net is publicly available. The paper also discusses related work in CT pancreas segmentation and attention mechanisms, and provides a detailed methodology for the proposed model, including its integration into the U-Net architecture and evaluation on challenging datasets.
Reach us at info@study.space
[slides] Attention U-Net%3A Learning Where to Look for the Pancreas | StudySpace